I confess I was a complete skeptic and luddite on the subject at the beginning of 2026. Totally missed the boat. Yet since then I have rapidly moved through a series of stages of acceptance. Initially resisting the reality of the situation but finally giving way to a new world order. Each state tollgated by the preceding state however, I do not believe I could have jumped straight to my current way of thinking.

Lets briefly marvel at the overarching profundity of what we have witnessed. I feel I need to write it down because I already take it for granted. It is already a part of my every day worklife. In January this year I thought, in my naivety, that intelligence was a biological phenomenon which certain lifeforms were imbued with. The ability to understand concepts, absorb the implications and then communicate with another intelligent beings was something that was 100%, purely and solely within the grasp of human beings. And yet, AI does this today, and nobody questions it any more.

How!!!??? Before I was comfortable using the technology (and I'm still not 100%) I had to try my best to answer this question. Using something I cannot begin to comprehend would be too much of a reach. Since it was just software & modelling after all, I felt I should understand it. Unlike hardware, where I never seem to worry about such details. Much reading of Wikipedia ensued to help understand how this new intelligence actually works.

Self-learning

Nobody spends time telling an AI this is an apple and this is a dog. Everything is learned by training on a huge set of written text - corpora. A model of understanding is built by slowly building an picture of how terms relate to one another. I could use the word 'predecessor' in three different sentences and if you did not know it before, by the end, you would have a good understanding of the concept. And let's face it (need go a bit Zen here...), everything IS just a concept and it's relationship to other concepts. Our, human interpretation is just one way, but it turns out there are other ways. The key feature is that it is self-learning approach, where simply (!) feeding in text to the process results in an understanding. Any text. This is what NLP (natural language processing) has been refining for years.

Embedding & Vectors

Words are ultimately converted into numbers. It had to be number-based, right? So that computers could handle it. No point in trying to mimic the biological processes of brain, whatever they are. But words are converted to vectors, which places the word in a multi-dimensional space of many co-ordinates. This space can have hundreds of dimensions, so the fidelty and subtleness of expression is extremely refined.

This is an 'embedding' process. Once embedded, there is a geometric relationship between terms which imply relative meaning between concepts. As per the common example, if you knew the concepts of London and the UK in your model, and someone asked you the capital of France, at the correct relative locality in the geometric space would be something called 'Paris' and it would be possible infer a good guess. Like a mixture of triangulation or interpolation.

From my understanding, a small team at Google lead the way on converting words to vectors (word2vec), in 2013.

This and more times a million

To go from these basic concepts that work in a lab, to products that are now on my computer and assisting me with a deep understanding of my work objectives is astounding. Colleauges are now talking about how to get the most of it. Or whether one model is marginally better than the other. Or how to write the best skill or what to put in your MCP endpoints. We have all moved from not expecting anything useful from it to fully embracing the capabilities. One small human phenomenon I have observed is how publically many will scorn AI but they are all using it.

There is so much more I want to understand about how it works. Like 'feed forward transformers' and how AI is able to move from 'understanding' to an interactive chat response that is in plain English and so meaningful. This is for future reading.

Is there still room for the skeptic?

Absolutely. I am confessing my disbelief at the power and utlity of AI today. It is a game changer, but it is still totally on my shelf with other tools which I use as and when I choose to. I'll never use it to help me write any code outside of work (and I write a fair bit). Or to help me write a blog (which explains the low quality!). Neither will I use it give me a recommedation on where to go or what to do. All these things are enjoyable elements of life that involve some creativity or some unpredictable connections. I don't see me ever deciding to delegate that.

However AI is increasingly difficult to avoid, so can I still choose? Now that when we search for something on the internet and the first hit is the AI explanation. I expect I will have to more selective about the tools we use.

Five Stages of AI acceptance

My stages of acceptance were as follows:

1. Object of Derision

Pre summer 2025

AI was mystical and fantastical concept that was just never going to happen. Joked with my colleagues about how one day we wouldn't have to do anything. Laughed at one guy who was always talking about how great it was going to be. Amused by my son's sudden enthusiasm about Chat GPT.

2. Ignorance & Rejection

Q3/Q4 2025

Still ignoring it. Actively rejected any talk of using it in work. Had to be coralled by a senior manager to install Copilot. Laughed when I heard people using Chat GPT when on holiday to get suggestions on where to go. Foolishly confident I could see out my career without dealing with it. It was an issue for the next generation.

3. Reluctant Use

Q1 2026

Found it useful for writing unit tests in work, saved me a lot of time, using just Copilot at this stage. Still wrangling with the fact it had to be trained on so much code written by people who did not consent. Ditto for internet chat models trained on what, books, newspapers, blogs ?!. Still largely unimpressed and surprised to see AI stocks flying

4. Mindblowing Discovery

Q2 2026

I get Claude Code in early access at work. Absolutely astounded at what it can do. Converted 8 Java classes to Rust in a few minutes for a fee of $0.57. It did it perfectly and would have taken me a day or two to do, at a cost of one hundred times this. Start writing skills to automate tedious repetitive tasks. AI stock story now makes sense but I am too late to benefit.

5. Adoption & Acceptance

Q3 2026 onwards

Having progressed from worrying about AI replacing parts of my job that I liked, to realising it is a workplace revolution that cannot be resisted. It is just too good. But fully expect there to be a payback. Companies will want to make cost cuts to make way for token spends, we can all join the dots. Starting to accept what it means for my career. Planning my way forward. Coding outside of work still, which is nothing new, but now more important.